Spectral Distance-Based Salience Feature-Aided Correlation Filter for Small Object Tracking in Hyperspectral Images

Object tracking using multi-band hyperspectral images (HSIs) has attracted a great deal of attention in recent years due to its potential to describe not only the temporal and spatial information of targets but also their material information. Accurately tracking small objects in low-light or low-contrast conditions remains challenging when using HSIs. In this paper, an environmental residual-aware multi-regularized correlation filter (ER-MRCF) is proposed by introducing environmental residual regularization into the objective function of MRCF, which improves the robustness of the CF tracker to environmental variations. Considering that the joint exploitation of texture features in false-color images and spectral features in HSIs helps improve the discrimination ability of CF-based trackers, spectral distance-based salience features are constructed and fused with spatial features to represent the target of interest in HSIs. Furthermore, a multipeak-aware module is adopted to evaluate the confidence of the tracking results and avoid model template contamination. Experiments conducted on the HOTC2024-RedNIR public dataset reveal that the proposed tracker achieves superior performance compared to several state-of-the-art CF trackers in terms of accuracy and success rate.